Improve GROUP BY on columns that are sorted

[smola]

We have some cases where data has some notion of order. For example, AFAIK in gitbase repositories, all rows with the same repository_id will be together, so it's possible to output its result as soon as the repository_id changes.

Taking this into account, a lot of aggregations do not need to keep full results in memory.

[erizocosmico]

That is only true if parallelism is 1, otherwise rows are just returned as they come from several partitions.

[smola]

@erizocosmico Then we would need info about the particion each row cromes from 😕